This paper proposes a random forest and adaptive stochastic ranking based multi-objective evolutionary algorithm(RFASRMOEA) for addressing expensive constrained multi-objective discrete optimization problems. To enhance the approximation accuracy of the surrogate model for discrete problems, the RFASRMOEA employs a random forest as the surrogate model to assist the evolutionary algorithm in the search process. Additionally, to improve overall performance, the algorithm introduces an adaptive stochastic ranking mechanism based on a balanced fitness evaluation strategy and adaptive probability operations. Specifically, the balanced fitness evaluation strategy uses population iteration information and incorporates diversity evaluation based on objective transfers and convergence evaluation based on cosine similarity, thus fully exploiting the potential of individuals in the population. The adaptive probability operations dynamically adjust the focus of the stochastic ranking mechanism, allowing the algorithm to explore a wider feasible domain in the early stage and then rapidly converge to the feasible domain in the later stage, thereby balancing the trade-off between satisfying the constraint conditions and optimizing the objective functions. Experimental results on test problems demonstrate that the proposed algorithm exhibits high competitiveness in dealing with expensive constrained multi-objective discrete optimization problems.